comparison of classic discriminant analysis and two-state logistic regression in separation of sardasht city, iran, chemical victims to mustard gas exposed and non-exposed groups

aims the aim of this study was to separate the chemical victims of mustard gas into exposed to and non-exposed groups to sulfur mustard using classical discriminant analysis and two-state logistic regression and selection of the best analysis. instruments methods the present study is a historical group that was conducted from 2005 to 2014. by observation method and systematic sampling, 284 people were included in the study including 216 people from sardasht city as an exposed group and 68 people from rabat city as a control group who were in all respects compared to the case group. using classical discriminant analysis and logistic regression methods, 32 quantitative variables were examined and finally these two methods were compared using rock curve analysis. spss 21 software was used for analysis. findings the 8 significant variables that had the highest ability to differentiate the groups (fev1/fvc ratio, testosterone, cholesterol, phosphorus, conjugated bilirubin, red blood cell count, hemoglobin and hematocrit) were selected and entered into the main models. using the rock curve, the cutting points of the variables were determined and the sensitivity and specificity values for discriminant analysis were 78% and 77.5%, respectively, and its sub-curved surface was 81.2%. in differentiating the groups, testosterone index was the strongest variable and conjugated bilirubin factor was the weakest variable. in logistic regression model, fev1/fvc, testosterone and phosphorus ratio variables were significant (p 0.05). the sensitivity and specificity of this model were 80% and 75%, respectively, the rock curvature level was 81.4% and the value of r^2 was 0.308. conclusion in the separation of chemical victims, the classical discriminant analysis and logistic regression methods have similar results, but discriminant analysis is a more appropriate model due to the presentation of the diagnostic function.